minimax-m3.1IMAGE
MiniMax M3.1 Flash 开源权重上 Hugging Face：原生多模态 MoE
HuggingPapers 称 M3.1 Flash 权重已上 Hugging Face，是面向编程、agent 与 computer use 的原生多模态 MoE，1M 上下文、全程稀疏注意力。
DailyPapers
@HuggingPapers5 天前收录原推发布于 2026.10.23
DailyPapers @HuggingPapers原推发布于 2026.10.23
MiniMax just released their M3.1 Flash model on Hugging Face A native multimodal MoE for coding, agentic workflows, and computer use with 1M token context and sparse attention throughout.
x.com/huggingpapers/status/2113437161999471090采集于 2026.10.23 12:09